AI & Data
Because the string in your index and the string in the search box are different strings. Measured on 297 realistic query variants against 47 documents in SQLite FTS5: a raw index answered 51.5%, a four-step normalisation pipeline answered 84.2%, and an exact-key lookup answered 15.8%. Plus the Vietn
AI & Data
Yes — measured 0.010108 against a predicted 0.010039 at one million keys, and no configuration was off by more than 5%. The formula is not the problem. Overfilling to twice the design capacity took the rate to 15.7%, a sum-of-character-codes hash took it to 99.9%, and deleting a single key of 100,00
Games
Because `position += velocity * deltaTime` is not frame-rate independent. The same falling body ended 0.477 m apart at 60 fps versus 144 fps after 10 seconds, and 1.43 m apart at 30 versus 240. A fixed-timestep accumulator closed the gap to zero — and then failed in three ways nobody measures.
Integration
max-age — and it is the only one on the list that saves a round trip. Measured on Node 23: a 304 from an ETag saved 307,502 of 307,835 bytes on a 300 KB bundle but still took 53.7 ms on an emulated 50 ms link, against 0.00 ms for a max-age hit. Includes the ETag hashing cost (md5 is 4.5x slower than
Integration
LIMIT/OFFSET is only correct if the table stops changing while you read it. Paging 100,000 rows on Postgres 16 while 100 a second were deleted lost 1,092 rows and raised no error. Keyset lost none — and the tiebreak on id is not optional.
Games
Serialising a 50,000-entity save with JSON.stringify took 16.5 ms — one whole 60 fps frame. But the frame is the cheap problem. When v2 of the game loaded v1 saves, JSON was silently wrong on two of three schema changes and the packed binary was silently wrong on the fourth. Measured across JSON, gz
AI & Data
A 300-entry character n-gram profile identified 100% of our 72 test tickets at 100 characters of input and 95.8% at 20 characters, in 266 microseconds each, offline. Unicode script detection resolves only 33% of them on its own. Measured on nine languages, plus the unaccented-Vietnamese collapse and
Games
One of them called Math.sin. Across 200,000 inputs, Node 23 and Python 3.14 on the same M3 disagreed on sin for 4.40% of them and on tan for 41.16% — by one to three ULP. That is enough to put two clients a full world unit apart in 464 steps, 7.7 seconds of play. The same simulation using only + - *
AI & Data
On a 187-sentence test set, text.split('.') produced 263 boundaries where 186 exist — 101 of them wrong. Intl.Segmenter, already in Node, scores 0.939 F1 for one line of code, but it splits after "Dr." and no locale option stops it.
Games
Keep the flat array and narrow it. Dropping tile ids from Int32 to Uint8 took a 4096x4096 terrain map from 64 MB to 16 MB and made random reads 3.3x faster, while hand-rolled RLE came out 1.23x larger than the plain array once both were gzipped. Measured across dense, sparse and layered maps at thre
AI & Data
A SHA-256 checksum matched 0 of 6,000 known duplicate image pairs. A 63-bit pHash at the Hamming threshold everyone hardcodes matched 67.4% of them and returned 7.7 false positives for every true match. Per-transformation precision and recall for aHash, dHash, pHash and a colour histogram, plus the
Games
Stop running A* once per unit. On a 512x512 grid with 500 units heading to one goal, per-unit A* took 904 ms and a flow field took 5.17 ms — 175x — and the flow field was already ahead at six units. Measured across four map types, five heuristics, three open-set structures and the no-path worst case